Title:BlockGC: A Joint Learning Framework for Account Identity Inference on Blockchain with Graph Contrast

Abstract:Blockchain technology has the characteristics of decentralization, traceability and tamper proof, which creates a reliable decentralized transaction mode, further accelerating the development of the blockchain platforms. However, with the popularization of various financial applications, security problems caused by blockchain digital assets, such as money laundering, illegal fundraising and phishing fraud, are constantly on the rise. Therefore, financial security has become an important issue in the blockchain ecosystem, and identifying the types of accounts in blockchain (e.g. miners, phishing accounts, Ponzi contracts, etc.) is of great significance in risk assessment and market supervision. In this paper, we construct an account interaction graph using raw blockchain data in a graph perspective, and proposes a joint learning framework for account identity inference on blockchain with graph contrast. We first capture transaction feature and correlation feature from interaction graph, and then perform sampling and data augmentation to generate multiple views for account subgraphs, finally jointly train the subgraph contrast and account classification task. Extensive experiments on Ethereum datasets show that our method achieves significant advantages in account identity inference task in terms of classification performance, scalability and generalization.